Seminar "Profiling mobile traffic at the edge with efficient machine learning methods" - Paolo Dini, Centre Tecnològic de Telecomunicacions de Catalunya (CTTC)

Wednesday, May 14th, 2025 at 14.30 in Aula Buzano DISMA
Prof. Paolo Dini, Centre Tecnològic de Telecomunicacions de Catalunya (CTTC)

Profiling mobile traffic at the edge with efficient machine learning methods

Abstract

The increasing demand for efficient resource allocation, enhanced security and privacy preservation in mobile networks has catalyzed the exploration of innovative solutions that could enhance the task of real-time mobile traffic modelling. Machine learning (ML) methods have become increasingly used to tackle traffic prediction in mobile networks, offering an attractive balance between the accuracy of the generated predictions and the timescale on which these are delivered when compared to alternative approaches such as analytical models or network simulators. In particular, deep learning (DL) has received significant attention due to its ability to capture the dynamics of network traffic both in terms of spatial and temporal dependencies across sites. While DL can provide highly accurate predictions for network traffic forecasting, its use is often compute-intensive, particularly when it requires substantial training datasets to achieve the desired accuracy. In this talk, we will explore solutions leveraging edge computing to alleviate computations of DL models for mobile traffic profiling. We will analyze multiple use-cases including traffic classification, prediction and anomaly detection. Moreover, we will discuss on possible future research directions to further increase the computational efficiency of ML when applied to mobile traffic profiling.